If demand doesn't fall down or current manufacturers supply go up, somebody (presumably in China) will spin up fabs. Apple wanted to use blacklisted Chinese RAM already.
I'm not sure if it's has any real bearing on real-world performance, but technically next token prediction makes it an online algorithm and they can be provably worse than (good) offline algorithms.
I haven't used windows in a long time, but on Android there's a dialog to pick browser and search engine on setup. So I'd assume that they're also bound by DMA in the same way.
On the other hand you can scale production of reactor themselves. And I don't think the idea is to scatter them around, but to have a power plant with dozens of them in one place (instead of 3-4 regular reactors in a regular nuke power plant).
And then USA will have a disadvantage compared to rest of the world with cheaper LLMs and american AI companies will have tougher time surviving on domestic spend alone.
I wonder how's model improvement/dollar invested ratio going. If gains are made by simply spending increasingly vast amounts of cash that's not going to last.